1
0
mirror of https://github.com/newnius/YAO-scheduler.git synced 2025-12-12 23:36:44 +00:00

[WIP] add distributed job support

This commit is contained in:
2019-06-05 17:09:22 +08:00
parent 6045e91cca
commit 9dd9c50002
3 changed files with 63 additions and 0 deletions

View File

@@ -8,6 +8,7 @@ import (
"io/ioutil"
"encoding/json"
"fmt"
"strconv"
)
type JobManager struct {
@@ -21,6 +22,8 @@ func (jm *JobManager) start() {
log.Println("start job ", jm.job.Name)
jm.jobStatus = JobStatus{Name: jm.job.Name, tasks: map[string]TaskStatus{}}
network := allocator.acquireNetwork()
/* request for resources */
for i := range jm.job.Tasks {
var resource NodeStatus
@@ -48,6 +51,9 @@ func (jm *JobManager) start() {
v.Set("name", jm.job.Tasks[i].Name)
v.Set("workspace", jm.job.Workspace)
v.Set("gpus", strings.Join(GPUs, ","))
v.Set("mem_limit", strconv.Itoa(jm.job.Tasks[i].Memory)+"m")
v.Set("cpu_limit", strconv.Itoa(jm.job.Tasks[i].NumberCPU))
v.Set("network", network)
fmt.Println(v.Encode())
@@ -102,6 +108,8 @@ func (jm *JobManager) start() {
time.Sleep(time.Second * 10)
}
allocator.releaseNetwork(network)
jm.allocator.finish(&jm.job)
log.Println("finish job", jm.job.Name)
}